Analytical Procedures for Determining the Impacts of Reliability Mitigation Strategies (2012)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

221 A p p e n d i x d This appendix describes the variables that are present in the data sets developed as part of the analysis of congestion causes performed by the University of Washington’s Washington State Transportation Center (TRAC-UW). This work was performed as part of the SHRP 2 L03 project. As noted in the main report, raw data were obtained from a variety of sources. Time, date, and location information (state route and mile post) were used to combine the various data items. Data were stored in a flat file record format, where each record in a file represents all data present for a specific five- minute interval in the year 2006 for a given direction for a given study segment. Consequently, each file contains 105,120 records of data. Because a separate file is used for each direction for each study corridor, there are 42 of these summary files produced for the SHRP 2 L03 project. The primary data storage and analysis system was Micro- soft Excel. (This effort is compatible with the 2007 version of Microsoft Office or any version thereafter, because the num- ber of records present in each file exceeds the allowable limit for earlier versions of Excel.) For a wide variety of analyses, these records also were read into various statistical packages (SPSS, SAS, and R), which allowed efficient computation of statistical tests. While a more capable database management system would be far more useful in the long term, the use of Excel allowed far easier development, testing, and analysis of derived statis- tics. Many of the statistics present in the analysis database are dependent on one or more data items from one or more prior time periods on that roadway segment. In Excel it was relatively easy to create these variables, test the variables, and visually examine how the variables reacted to changing traffic condi- tions (e.g., high/low volume and high/low speeds). It also was possible to easily find and examine how new test variables changed over time, given multiple different secondary inputs. It also is easy to identify specific anomalies (e.g., time periods with large amounts of congestion, but no traffic disrup- tions noted by a newly computed test variable, and track the performance of that computation over time). This process allowed the research team to identify specific computational techniques that did not work consistently. It also produced a better analysis database. Note that the actual Excel computational formulas are not all included in these final datasets. Including all of the com- putations causes Excel to exceed the number of computations allowed in a single file. This causes unstable behavior within Excel. Consequently, once a computational variable was deter- mined to work as intended, the data resulting from that latest series of computations was converted from an active Excel for- mula (i.e., recomputed each time variables were recomputed within Excel) to a constant. This was normally accomplished by simply saving the dataset as a CSV file and reimporting those values into a new Excel file. In other cases, especially cases where very complex, logical processes were necessary to compute new row values, a separate computational spreadsheet was used to produce one or more new columns of data. These were then cut-and-pasted into the primary analysis spreadsheets. The following variables can be found in the final data sets developed and used in the L03 project by TRAC-UW (see Table D.1). In some datasets, specific variables were not com- puted. When this occurs, the term empty is included in the variable name column, indicating that this variable does not exist for the spreadsheet being examined. Queue Extended Influence Mathematically, the queue extended influence method assumes that any increase in travel time that occurs while an incident is active is associated with that incident. That is, if travel times increase over the travel time experienced at any time during an incident, that longer travel time is caused (at least in part) by the incident, even if other events are occurring in the corridor. An incident is defined as being active in two ways: 1) the incident is actually recorded as taking place within that specific five-minute period; and 2) a trip that entered Seattle Analysis: Variable Definitions

222 Table D.1. Data Items Used in the Seattle Congestion-by-Source Analysis Column Variable Name Definition A link_name Roadway and Direction. B direction East, West, North, or South. C Date Date (M/D/Yr). D Time 0:00 to 23:55 in 5-minute increments. E Decimal Time Time expressed as a decimal. F Hour The hour of the day. G Day of Week Numeric day with 1 = Sunday to 7 = Saturday. H Month Numeric month with 1 = January to 12 = December. I Day Numeric day from 1 to 31 of each month. J [route]_[segment]_TT[dir] Travel Time [direction] on [route], [segment] section. K avg_occupancy Average Occupancy from Operation Archive. L avg_vht Average VHT from Operation Archive. M avg_volume Do not use, not a good value. N Accident Severity Severity of Accident (1 = PDO, 2 = injury, 3 = fatal). O Accident Accident Variable, equal 1 when an accident occurred. P Accident Severity (Calc) The total number of 5-minute time periods during which a queue, influenced by a given accident lasts. (Value exists only for the first 5-minute period during which the accident occurs.) Q Max_closure_length Maximum duration closure of lane(s) from Operation Archive. R Closure Severity The total number of 5-minute time periods during which a queue, influenced by a given lane closure lasts. (Value exists only for the first 5-minute period during which the closure occurs.) S Max_incident_length Maximum duration of incident from Operation Archive. T Incident Severity The total number of 5-minute time periods during which a queue, influenced by a given incident lasts. (Value exists only for the first 5-minute period during which the incident occurs.) U Accident Rubbernecking Has a value of 1 whenever there is an accident on the other side of the road. V Incident Rubbernecking Has a value of 1 whenever there is an incident on the other side of the road. W Delay Variable Computed Vehicle delay (Actual Travel Time—Free Flow Travel Time) * Maximum Section Volume. X IF Variable Variable describing event effects present during that 5-minute time period: 0. No cause of congested noted in available variables; 1. ONLY Acc Queue Extended is present; 2. ONLY Inc Queue Extended is present; 3. ONLY Precipitation hour is present (it has rained in the past hour); 4. BOTH Acc Queue Extended and Inc Queue Extended are present; 5. BOTH Acc Queue Extended and Precipitation hour are present; 6. BOTH Inc Queue Extended and Precipitation hour are present; and 7. All three variables are present. Note: This variable is based on the 5+15 queue extended methodologya Y Max_occupancy Maximum Occupancy from Operation Archive. Z Max_speed Maximum Speed from Operation Archive. AA Max_volume Maximum Volume from Operation Archive. AB Min_speed Minimum Speed from Operation Archive. AC Min_volume Minimum Volume from Operation Archive. (continued on next page)

223 AD Sum_vht Sum of the VHT from Operation Archive. AE Sum_vmt Sum of the VMT from Operation Archive. AF Accident -5+20 Associated with Accident variable (Column O)—time periods back 5 minutes and ahead 15 minutes from an accident (the variable is slightly misnamed). AG Acc Queue Extendeda Associated with Accident variable (Column O)—time periods extended from accident time period according to queue extended method (see end note). AH Inc -5+20 Associated with Max_incident_length variable (Column Q). AI Inc Queue Extended Associated with Max_incident_length variable (Column Q). AJ Closure -5+20 Associated with Max_closure_length variable (Column P). AK Closure Queue Extended Associated with Max_closure_length variable (Column P). AL Acc + Closure -5+20 Associated with the combination of Accident and Max_closure_length variables (Column O and P). AM Acc + Closure Queue Extended Associated with the combination of Accident and Max_closure_length variables (Column O and P). AN Acc + Inc -5+20 Associated with the combination of Accident and Max_incident_length variables (Column O and Q). AO Acc + Inc Queue Extended Associated with the combination of Accident and Max_incident_length variables (Column O and Q). AP Acc + Inc + AccRub -5+20 Associated with the combination of Accident, Max_incident_length and Accident Rubbernecking variables (Column O, Q, and R). AQ Acc + Inc + AccRub Queue Extendeda Associated with the combination of Accident, Max_incident_length and Accident Rubbernecking variables (Column O, Q, and R). AR Acc. + Inc. + Rub -5+20 Associated with the combination of Accident, Max_incident_length, Accident Rubber- necking and Incident Rubbernecking variables (Column O, Q, R, and S). AS Acc. + Inc. + Rub Queue Extended Associated with the combination of Accident, Max_incident_length, Accident Rubber- necking and Incident Rubbernecking variables (Column O, Q, R, and S). AT Space Mean Speed Average Speed derived from Travel Time and Segment Length. AU Rounded Speed 5.0 Rounded Average Speed (Column AT) to the nearest 5.0 mph. AV Rounded Speed 2.5 Rounded Average Speed (Column AT) to the nearest 2.5 mph. AW Rounded Speed 2.0 Rounded Average Speed (Column AT) to the nearest 2.0 mph. AX Regime The condition of the road segment (minimum speed observed and maximum volume observed) (1 = lots of capacity left, 2 = less than one lane of capacity, 3 = minimal capacity left, speed slowed slightly, 4 = congestion present, 5 = recovery underway). AY Holiday Has a value of 1 on the following days: Jan 2, Feb 20, May 29, July 3, July 4, Sep 4, Nov 10, Nov 23, Nov 24, Dec 25, Dec 26. AZ Rain 1 if NOAA Weather Type of Rain(RA), Mist(BR), Drizzle(DZ), T-storm(TS), or Haze(HZ) for the most recent time period reported (0 otherwise). BA Heavy_Rain 2 if Rain as defined above with NOAA hourly precipitation > 0.125 inches (0 otherwise). BB Wind 3 if NOAA Wind speed greater than 19 mph (0 otherwise). BC Snow 4 if NOAA Weather Type of Snow(SN), Freezing(FZ), Sm Hail(GS), Hail(GR), Ice Pellet(PL) or Squall(SQ) for the most recent time period reported (0 otherwise). BD Fog 5 if NOAA Weather Type of Fog(FG) OR NOAA Visibility < 0.25 (0 otherwise). BE Wind-Speed Wind speed (in knots) directly from NOAA data for the most recent time period reported. BF Wind-Gusts Wind speed for gusting winds (in knots) directly from NOAA data for the most recent time period reported. Table D.1. Data Items Used in the Seattle Congestion-by-Source Analysis (continued) Column Variable Name Definition (continued on next page)

224 BG Precip_hour Hourly precipitation (in inches and hundredths) from the most recent reported hourly NOAA data. BH Precip_2hours Sum of last 2 hours of precipitation. BI Precip_4hours Sum of last 4 hours of precipitation. BJ Precip_8hours Sum of last 8 hours of precipitation. BK Hours_since_rain Number of hours since last reported precipitation of any amount. BL R2-R4_5 min Tells us that there was a change from Regime 2 to Regime 4 within the last 5 minutes. BM R2-R4_10 min Tells us that there was a change from Regime 2 to Regime 4 within the last 10 minutes. BN R2-R4_15 min Tells us that there was a change from Regime 2 to Regime 4 within the last 15 minutes. BO R3-R4_5 min Tells us that there was a change from Regime 3 to Regime 4 within the last 5 minutes. BP R3-R4_10 min Tells us that there was a change from Regime 3 to Regime 4 within the last 10 minutes. BQ R3-R4_15 min Tells us that there was a change from Regime 3 to Regime 4 within the last 15 minutes. BR Number_of_2ndary_events_Accidents Uses the Severity (duration) variable and then looks to see how many accidents and incidents occur within the duration (time the queue is present) of the accident in question. BS Numb_Sec_Rubnking_Accidents Uses the Severity (duration) variable and then looks to see how many accident and inci- dent rubbernecking events occur within the duration (time the queue is present) of the accident in question. BT Number_of_2ndary_events_Closures Uses the Severity (duration) variable and then looks to see how many accidents and incidents occur within the duration (time the queue is present) of the closure in question. BU Numb_Sec_Rubnking_Closures Uses the Severity (duration) variable and then looks to see how many accident and inci- dent rubbernecking events occur within the duration (time the queue is present) of the closure in question. BV Number_of_2ndary_events_Incidents Uses the Severity (duration) variable and then looks to see how many accidents and incidents occur within the duration (time the queue is present) of the incident in question. BW Numb_Sec_Rubnking_Incidents Uses the Severity (duration) variable and then looks to see how many accident and inci- dent rubbernecking events occur within the duration (time the queue is present) of the incident in question. BX 5+5 Queue Extended Crash The queue extended variable (1 = influence is present) using the 5-minute follow on period as the basis for computation crashes only. BY 5+5 Queue Extended Incident The queue extended variable (1 = influence is present) using the 5-minute follow on period as the basis for computation incidents only. BZ 5+5 Queue Extended Closure The queue extended variable (1 = influence is present) using the 5-minute follow on period as the basis for computation closures only. CA 5+5 Queue Extended Rubbernecking The queue extended variable (1 = influence is present) using the 5-minute follow on period as the basis for computation either rubbernecking variable is active. CB 5+5 Queue Extended Incident or Accident The queue extended variable (1 = influence is present) using the 5-minute follow on period as the basis for computation if an incident or accident has occurred. CC Mainline IF Variable 5+5 Sets a value 1-8 (see column X definition for what each value means) indicating what influences are present to cause congestion. Examines only WITHIN segment variables—and does NOT include construction effects. This version of the “IF” variable is based on the 5+5 Queue Extended computations and the variables in columns BX through CB. Table D.1. Data Items Used in the Seattle Congestion-by-Source Analysis (continued) Column Variable Name Definition (continued on next page)

225 CD Construction variable 1. Construction. 2. All lanes closed. 3. 520 weekend closures. 4. Construction happens in two locations at one time in the same segment. CE IF 5+5 Construction Included Variable describing event effects present during that 5-minute time period. Uses the 5+5 Queue Extended variables as input AND includes notifications of construction traffic management activities. 0. No cause of congested noted in available variables; 1. ONLY Inc Queue Extended is present; 2. ONLY Acc Queue Extended is present; 3. ONLY Precipitation hour is present (it has rained in the past hour); 4. BOTH Acc Queue Extended and Inc Queue Extended are present; 5. BOTH Inc Queue Extended and Precipitation hour are present; 6. BOTH Acc Queue Extended and Precipitation hour are present; 7. All three variables are present; 8. Ramp congestion, but no cause for ramp congestion is known; 9. Construction activity going on; 10. Construction activity plus ramp congestion; 11. Construction activity plus an incident queue extended; 12. Construction activity plus an accident queue extended; 13. Construction activity plus rain; 14. Construction activity plus an accident and incident queues extended; 15. Construction activity plus an incident queue extended and rain; 16. Construction activity plus an accident queue extended and rain; and 17. Construction activity plus an accident and incident queues extended and rain. CF Delays caused by ramps/downstream queues (1st location) A nonzero value is present when loop detectors at a ramp have lane occupancy greater than 35%. (This is used as a measure that queues have formed on the ramp and are likely to cause congestion on the connecting roadway.) Uses the same variable defini- tions as in column CE. The name in the header row changes from dataset to dataset to describe the specific ramp and/or downstream segment. There are three columns allocated for these external to the road segment variables CF, CG, and CH. CG Delays caused by ramps/downstream queues (2nd location) See CF definition. CH Delays caused by ramps/downstream queues (3rd location) See CF definition. CI IF—Single Cause (5+5) Combines the causes defined in the variables in CE, CF, CG, and CH. The effects are cumulative. So that a “1” on the mainline and a “2” on a connecting ramp means this variable would become a “4” (both accident and incident effects). CJ Rounded Converts the Time variable to half hour increments (0 for 0:00 through 0:25, 0.5 for 0:30 through 0:55, 1 for 1:00 through 1:25) to allow easy aggregation of results on a half hour basis. CK Crash versus Volume Is a three category variable. The variable is set to 0 when no known disruption is affect- ing roadway performance. It is set to the value “1” when a crash is affecting roadway performance. It is set to a “2” when some other (noncrash) is influencing roadway performance. (The value is “1” when a crash influences performance, even if other factors also influence that performance.) CL Incident versus Volume Is similar to the Crash versus Volume variable, except that the value “1” is used to indi- cate that an incident reported by WSDOT’s incident response team is influencing road- way performance. A “2” indicates some disruption other than something reported by WITS is influencing roadway performance. CM Queue Duration Incidents The number of 5-minute time periods during which the roadway is influenced (traffic is slower than the fastest travel time observed during an incident) for a defined incident. One value exists for each incident for which there is a valid travel time. That value is placed in the row that corresponds to the first occurrence of the incident. Table D.1. Data Items Used in the Seattle Congestion-by-Source Analysis (continued) Column Variable Name Definition (continued on next page)

226 CN Queue Duration Crashes The number of 5-minute time periods during which the roadway is influenced (traffic is slower than the fastest travel time observed during a crash) for a defined crash. One value exists for each crash for which there is a valid travel time. That value is placed in the row that corresponds to the first occurrence of the crash. CO Queue Duration Closures The number of 5-minute time periods during which the roadway is influenced (traffic is slower than the fastest travel time observed during an incident) for a defined incident involving a lane closure. One value exists for each closure for which there is a valid travel time. That value is placed in the row that corresponds to the first occurrence of the closure. CP Queue Duration Incidents and Crashes The number of 5-minute time periods during which the roadway is influenced (traffic is slower than the fastest travel time observed during an incident) for any defined inci- dent or crash. One value exists for each incident or crash for which there is a valid travel time. That value is placed in the row that corresponds to the first occurrence of each incident or crash. CQ Rubbernecking Influence Duration 5+5 The number of 5-minute time periods during which the roadway is influenced (traffic is slower than the fastest travel time observed during a rubbernecking event) for a defined rubbernecking event. One value exists for each rubbernecking event for which there is a valid travel time. That value is placed in the row that corresponds to the first incidence of the rubbernecking event. CR When Congestion Ends a.m.b This variable places a “1” in the first row which defines a noncongested condition during the A.M. peak period. “Not Congested” is defined as being four consecutive rows where travel times are faster than 1.05 times travel at the speed limit (i.e., faster than 57.15 mph.) For the a.m. time period, this event cannot take place prior to 7:00 a.m. It can occur any time AFTER 7:00 a.m. The row selected is the FIRST row in which the four consecutive rule is observed. Congestion due to a late occurring incident may cause congestion after this occurrence. This congestion is ignored by this variable. CS Incident Effected a.m.—4:00 a.m. Start If ANY incident occurs after 4:00 a.m. on a given day, this variable is set to “1” at the time the first incident occurs. It remains set to “1” for the rest of the day. CT Crash Effected a.m.—4:00 a.m. Start If ANY crash occurs after 4:00 a.m. on a given day, this variable is set to “1” at the time the first crash occurs. It remains set to “1” for the rest of the day. CU Selection a.m. The section variable is set to “1” if the day is a Tuesday, Wednesday, or Thursday, AND it is not a designated holiday AND the “When Congestion Ends a.m.” variable is set to “1”. CV Effected Inc versus Crash versus Nothing a.m. This categorical variable is set to “0” unless: a crash has occurred (value = 1) or an inci- dent has occurred (value = 2). When both an incident and crash have occurred, the value is set to “1”. CW When Congestion Ends p.m. This variable places a “1” in the first row which defines a “noncongested” condition during the P.M. peak period. “Not Congested” is defined as being four consecutive rows where travel times are faster than 1.05 times the travel time at the speed limit (i.e., faster than 57.15 mph.) For the P.M. time period, this event cannot take place prior to 4:00 p.m. It can occur any time AFTER 4:00 p.m. The row selected is the FIRST row in which the four consecutive rule is observed. Congestion due to a late occurring incident may cause congestion after this occurrence. This congestion is ignored by this variable. CX Incident Effected p.m.—4:00 p.m. Start If ANY incident occurs after 3:00 p.m. on a given day, this variable is set to “1” at the time the first incident occurs. It remains set to “1” for the rest of the day. CY Crash Effected p.m.—4:00 p.m. Start If ANY crash occurs after 3:00 p.m. on a given day, this variable is set to “1” at the time the first crash occurs. It remains set to “1” for the rest of the day. CZ Selection p.m. The section variable is set to “1” if the day is a Tuesday, Wednesday, or Thursday, AND it is not a designated holiday AND the “When Congestion Ends p.m.” variable is set to “1”. DA Effected Inc versus Crash versus Nothing p.m. This categorical variable is set to “0” unless: a crash has occurred (value = 1) or an inci- dent has occurred (value = 2). When both an incident and crash have occurred, the value is set to “1”. a Queue Extended Influence assumes that once an incident has occurred, any travel time increase along the corridor is associated with the (potential) queue that forms as a result of that incident, and thus all travel in the corridor is affected by that incident until the queue has fully dissipated. b A discussion concerning the “When Congestion Ends” variable—Considerable testing went into the selection of the time period at which congestion was described as ending. Table D.1. Data Items Used in the Seattle Congestion-by-Source Analysis (continued) Column Variable Name Definition

227 the test section in the time period immediately prior to that incident occurring or immediately after that incident stopped occurring. These time periods are indicated as 5+5 in the variable names in the study spreadsheet. They account for the fact that a trip starting at 7:00 but requiring 10 minutes to traverse a section, may be adversely affected by an incident which occurs during the 7:05 time period. Conversely, the queue caused an incident occurring and cleared in the 7:05 time period, might not grow to the point where the loop sensors used in this analysis notice that queue until the 7:10 time period. These additional 10 minutes were referenced internally as the time extended incident period. The queue extended incident period uses these measures of extended duration within which to find the base travel time against which continued queuing is measured. (See two paragraphs below for the defi- nition of how this process works.) The initial test of the time extension variable was 5 minutes prior and 15 minutes after an incident, PLUS the (minimum) 5-minute period containing the incident. This initial set of analyses was called 5+20. The variables created and used in these initial computations are still present in the data set and are stored in columns AF through AS. The 15-minute exten- sion was determined to be too lenient. That is, it was unclear that the delays beginning 15 minutes after the incidents had been cleared were related to the incident. This led to the adop- tion of the 5+5 rule. The 5+20 variables were not used in any of the published analyses, but have been left in the analysis data set to allow future analysis should they be of interest. The queue extended computation begins by determining the fastest corridor travel time experienced in the 5-minute period before the disruption occurs through 5 minutes after that disruption is reported to have ended. All subsequent travel periods are assumed to be influenced by that disruption until corridor travel times return to (are equal to or faster than) the fastest time observed during the “5+5” time period (5 minutes prior to the incident through 5 minutes after the disruption). By changing the definition of disruption and reapplying these basic rules, the influence of any combination of disruptions can be indicated. This approach does mean that for the ana- lytical purposes of this project—the influence of any disrup- tion lasts at least 15 minutes. In off-peak conditions (where low volume exists—or in other words there is considerable unused roadway capacity), this approach is an excellent measure of incident effects. If the incident occurs at the beginning of peak period conditions, the approach is likely to associate all of the peak-period con- gestion with the incident. This is assumed to be acceptable based on the concept that the incident condition combined with the growing peak-period traffic volume will cause con- gestion to form earlier than would otherwise have occurred on that particular peak period, and the increased congestion will cause travel times to remain elevated later into the tail end of the peak period. While this is a liberal measure of the congestion caused by a given incident, and may overstate the extent of any given incident’s congestion causing influence, it does replicate the “lasting influence” that an incident can have on roadway performance. An example of how the Queue Extended Influence Area works and why it is used is as follows. On Thursday, February 23, 2006 at 2:30 p.m., on SR 520 westbound headed into the city of Seattle, traffic is flowing slightly better than normal (travel time = 494 seconds versus an annual mean travel time of 549 seconds for the 2:30 p.m. time period.) A lane closing incident, which takes 17 minutes to clear, occurs. During that incident, travel times through the corridor slow to 738 seconds. After the incident is cleared, congestion begins to clear out, but does not return to its pre- incident condition, before increasing traffic volumes associ- ated with the start of the p.m. peak period cause travel times to again degrade. (That is, the queue formed by the incident has yet to fully clear and thus P.M.-peak-period congestion occurs earlier than normal, because the incident caused queue has reduced the roadway’s capacity, making it unable to serve the volumes associated with the beginning of the P.M. peak shoulder period.) Travel times in the P.M. peak are thus slower than normal throughout the peak period, and do not return to preincident conditions until well after the P.M. peak period ends. (Maximum travel time on this day in the traditional P.M. peak is 1,787 seconds at 6:05 p.m.) Before the P.M. peak ends, a two-car, injury collision occurs (at 6:50 p.m.) within the roadway analysis segment. Travel speeds (already bad) degrade considerably after the accident (reaching 2,960 seconds, or an average speed of 8.5 mph for the 7 mph road segment), and don’t return to preaccident conditions (still much slower than the original preincident conditions) until 8:05 p.m. However, once again, before the queue can fully dissipate, a second injury accident occurs at 8:35 p.m. Travel times again climb, despite the lower traffic volumes experienced at 8:30 in the evening, peaking this time at just over 1,900 seconds. Only after this accident and its resulting queue is cleared, do travel times finally return to a point faster than that found at 2:35 p.m., just prior to the first incident. This occurs at 10:10 p.m. Fur- ther contributing to congestion on this day are two other factors, 1) a higher travel demand than normal caused by a University of Washington’s men’s basketball game (the team was ranked 17th in the country at the time) which occurred that evening at the University basketball arena located at the west- ern end of this analysis corridor (the game was a 10,000 per- son sell out event, and started at 7:30 p.m.); and 2) the fact that it rained off and on that afternoon and evening. (The 6:50 p.m. accident notes rain and wet pavement conditions, while the 8:35 p.m. accident notes dry pavement and overcast conditions.)

228 The reality of this day is that a variety of events helped cause the congestion experienced by travelers. However, the instigating event appears to have been the original lane- blocking incident. Without that event and the extended queues it creates, it is possible that neither rear-end accident would have occurred. (Although it is impossible to directly tie the rear-end collisions with a specific queue length.) On the other-hand, delays would have been considerably smaller without the two accidents and without the added travel demand caused by the basketball game. Similarly, the rain may very well also have contributed to the cause of conges- tion on the corridor as well as the occurrence of both of the accidents, as well as to the time it took for the roadway to recover from all three events. Consequently, we believe that the queue extended influence variable works as defined— it indicates that a given event is likely to have influenced—but not totally caused—the level of congestion experienced on the roadway. The queue extended approach successfully tracks the exis- tence of that queue to that instigating event. What remains is to determine how to describe the relative importance of that event versus the contributions of the other causal factors. When Congestion Ends The project team was concerned that variations in speed both over time and over the length of the study section might give false indications that free flow travel had returned to the study section, when what was really being measured was a temporary improvement in conditions caused by random fluctuations in traffic density. Consequently, it was decided that the determi- nation of when congestion abated must include both the facts that speeds were free flow throughout the study section and that they remained so for long enough to ensure that the obser- vation was not just a temporary change in conditions. After testing the variation of travel times at the end of the peak peri- ods on multiple study sections, it was determined that 20 min- utes of consecutive travel times below a set value ensured that flow remained in a fluid state. However, while 20 minutes of fluid flow are required, the “end point” for congestion is indi- cated as the first of those 20-minute periods. The selection of the speed at which congestion ended was set based on the available data. The analysis data set had three measures of “speed”—maximum speed in the segment, minimum speed in the segment, and travel time through the segment. Travel time was selected as the variable of choice for two reasons: 1) on longer test sections, it more effectively replicates the travel conditions experienced by motorists—when compared to maximum and minimum speed values selected from differ- ent locations within that segment, but within a single time slice; and 2) travel time effectively accounts for the importance of different speed measurements along the length of the seg- ment, while minimum and maximum values can shift from one location to another from one time slice to another. Thus use of travel time moderates the importance of any one slow speed measurement, while the 20-minute requirement ensures that fluctuations in the observed travel times do not artificially cause the procedures to end congestion too quickly. Tests were made using 5%, 10%, and 20% increases in travel times, corresponding to average segment speeds of 57, 54, and 48 mph. Each of these travel time increases can be achieved in a variety of ways, ranging from modest slowing throughout the section, to a more substantial slowing at any one speed mea- surement location with a section. Tests of these speeds indicated that on most corridors, the slower travel times were frequently met during the middle of the traditional peak periods. As a result, they were assumed to be too lenient a travel time mea- sure. (That is, one location of moderate congestion—speeds below 40 mph—could occur while travel times remained fast enough to meet the 10% increase criteria.) Thus, it was neces- sary to select the more stringent criteria of only a 5% increase in travel times. When combined with the 20-minute requirement, this gave results which matched local experts’ general impres- sions in all cases except 11 corridors during the A.M. peak period. These corridors all experienced some level of routine vehicle slowing during the middle of the day, and thus fre- quently never reached “the end of congestion” as defined by the 5% and 20-minute rules until after the P.M. peak had ended. As a result, for those 11 routes, for the A.M. peak, either the 10 or 20% rules were applied in order to ensure that congestion “ends” prior to noon on days when no-incident occurs. The lowest percentage increase which ended mean travel time prior to noon for days in which no incident occurred was selected for each of these 11 corridors. (That is, if the mean “congestion end” time point for nondisrupted days occurred prior to noon based on a 10% increase in travel time that value was used. The 20% value was only used if the 10% value forced congestion to end after noon.)